OHBM: Want to hear more about Bayesian methods for analysing lesions?
I’m super excited to present some new additions that we made to BLESS (= Bayesian Lesion Estimation with a Structured Spike-and-Slab Prior) this year at OHBM 2022. With BB-BLESS (= Bayesian Bootstrap - BLESS), we add some really cool features to the principled analysis of lesions with Bayesian spatial regression modelling. Here is a list of a few of BLESS’s features:
- Scales to really large datasets.
- Handles many covariates and produces a spatial parameter map for each of them.
- Bayesian shrinkage priors to do variable selection for each covariate and voxel.
- Fast inference methods via optimization rather than MCMC (Variational Inference + Bayesian Bootstrap methods).
- Uncertainty estimates of any spatial statistics, such as cluster size.
- New reliability measures of cluster size via approximate posterior sampling.
- UK Biobank application with 40,000 subjects and 50,000 voxel locations.
If any of this sounds interesting, then please come by my poster #WTh036 titled “Bayesian Bootstrap Uncertainty Quantification for Spatial Lesion Regression Modelling” on Wednesday and Thursday (22nd and 23rd of June, 2022) between 1:45PM - 2:45PM GMT +1 and say hi! Always happy to arrange an online chat or a meet up at a different time during OHBM too. Just get in touch on Twitter or via Email :)
Here is a link to the full OHBM 2022 Poster.